WO2017216465A1

WO2017216465A1 - Method and device for processing images acquired by a camera of a motor vehicle

Info

Publication number: WO2017216465A1
Application number: PCT/FR2017/051510
Authority: WO
Inventors: Lucien Garcia; Arnaud FAURE
Original assignee: Continental Automotive France; Continental Automotive Gmbh
Priority date: 2016-06-14
Filing date: 2017-06-13
Publication date: 2017-12-21
Also published as: FR3052582A1; FR3052582B1

Abstract

The present invention relates to a method for processing images acquired by a camera of a motor vehicle. The method makes it possible to determine (E9) an epipolar segment in an image, to select (E10) a plurality of points on the determined segment, and, for each determined point, to calculate (E11A) the distance of a predetermined projection plane, to determine (E11B) an intermediate set of pixels by projecting a first determined set on the predetermined plane located at the distance calculated, and to calculate a correlation score for each set of pixels associated with a point of the determined segment in order to determine the pixel of the second image which represents the same point as the associated pixel in the first image.

Description

Method and device for processing images acquired by a camera of a motor vehicle

The invention relates to the field of assistance for driving a motor vehicle and more particularly relates to a method and a device for processing images acquired by a camera of a motor vehicle and a motor vehicle comprising a such device.

Nowadays, it is known to equip a motor vehicle with a driver assistance system commonly called ADAS ("Advanced Driver Assistance System" in English). Such a system comprises in known manner a video camera, for example mounted on the front windshield of the vehicle, which generates a flow of images representing the environment of said vehicle. These images are used by an image processing device for the purpose of assisting the driver, for example by detecting an obstacle or crossing a white line. The information given by the images acquired by the camera must therefore be sufficiently precise and relevant to allow the system to assist the driver reliably.

It is known to detect an object from a series of images acquired by the video camera by following the optical flow ("optical flow" in English) of a number of points in the series of images. More precisely, it is a question of following the trajectory of these points in a plurality of successive images and to determine their location in the space.

Some pixels representing the sky or the road are however difficult to track from one image to another because of their uniformity in several successive images, which has significant disadvantages.

In order to follow a point represented by a pixel of a first image in a second consecutive image, a known solution is to analyze an area of pixels around said pixel in the second image to determine in which direction the point is moving between the first image. picture and the second picture.

Such a solution may require the use of a pixel area comprising more than 2500 pixels, which requires significant processing capabilities and therefore has a major disadvantage for an embedded system in a motor vehicle.

In addition, when the vehicle is traveling at a high speed, for example more than 130 km / h, it is difficult or impossible to follow the points in successive images with such a method, including points representing objects in perspective.

Finally, such a solution, although robust, makes it possible to track the waves only a very limited number of points of an image, of the order of 2% of points, which can be insufficient to detect a significant number of objects and thus to effectively assist the driver in his conduct.

The invention therefore aims to remedy at least in part these disadvantages by providing a simple, reliable and effective solution for detecting objects in a video stream captured by the camera of a motor vehicle.

To this end, the subject of the invention is a method of processing images acquired by a camera of a motor vehicle, said method comprising the steps of:

Acquisition of a first image and a second image by the camera, determination of a rotation matrix and a translation matrix between the first image and the second image,

• selection of at least one pixel in the first image,

• for each pixel selected in the first image:

determining, in the first image, an initial set of pixels comprising the selected pixel and a plurality of pixels located in the vicinity of the selected pixel,

determination, in the second image, of a first set of pixels by projection on a predetermined plane situated at a predetermined minimum distance from the determined initial set, from the determined rotation matrix and translation matrix, said first set comprising a first pixel corresponding to the pixel selected in the first image,

determination, in the second image, of a second set of pixels by projection on the predetermined plane situated at a predetermined maximum distance from the determined initial set, from the determined rotation matrix and translation matrix, said second set comprising a second pixel corresponding to the pixel selected in the first image,

determining a segment connecting the first pixel and the second pixel, selecting a plurality of points on the determined segment,

- for each selected point:

Calculating the distance of the predetermined projection plane,

Determination, in the second image, of a set of intermediate pixels by projection on the predetermined plane situated at the calculated distance, of the initial set determined, from the determined rotation matrix and of the translation matrix, for each set of pixels associated with a point of the determined segment, calculating a correlation score,

determination in the second image, from the computed correlation score, of the pixel corresponding to the same object on the first image and on the second image, thus representing the optical flow of the same pixel between the two images.

The use of an epipolar segment makes it possible to improve the determination of the optical flow of a pixel between two consecutive images. This method makes it possible to obtain optical flows for a number of pixels that can reach at least 10% of the total number of pixels present in an image, which makes it possible to detect a large number of notable objects such as bridges, panels, poles, etc. The method according to the invention can advantageously be used on pixels representing both static and dynamic objects.

According to one aspect of the invention, the method further comprises a step of determining the minimum distance and the maximum distance.

According to one aspect of the invention, the method further comprises a step of determining the projection plane.

Preferably, the correlation score of a set of pixels associated with a point of the determined segment is calculated from the sum of the absolute differences between the gray levels of the pixels of said set and the gray levels of the initial set determined. such a calculation being easy to perform.

The invention also relates to an image processing device for a motor vehicle, said device comprising a camera capable of acquiring a first image and a second image and an image processing module configured for:

· Determining a rotation matrix and a translation matrix between the first image and the second image acquired by the camera,

• select at least one pixel in the first image,

• for each pixel selected in the first image:

determining, in the second image, a first set of pixels by projection on a predetermined plane situated at a predetermined minimum distance from the determined initial set, from the determined rotation matrix and translation matrix, said first set of pixels set comprising a first pixel corresponding to the pixel selected in the first image, determining, in the second image, a second set of pixels by projection on the predetermined plane situated at a predetermined maximum distance from the determined initial set, from the determined rotation matrix and translation matrix, said second set of pixels set comprising a second pixel corresponding to the pixel selected in the first image,

determining a segment connecting the first pixel and the second pixel,

select a plurality of points on the determined segment,

- for each selected point:

Calculate the distance, called the intermediate distance, from the predetermined projection plane,

• determining, in the second image, a set of intermediate pixels by projection on the predetermined plane located at the calculated intermediate distance, of the initial set determined, from the determined rotation matrix and translation matrix,

for each set of pixels associated with a point of the determined segment, calculating a correlation score,

determining in the second image, from the computed correlation score, the pixel corresponding to the same object on the first image and on the second image.

According to one aspect of the invention, the image processing module is configured to determine the minimum distance and the maximum distance.

According to another aspect of the invention, the image processing module is configured to determine the projection plane.

Preferably, the image processing module is configured to calculate the correlation score of a set of pixels associated with a point of the segment determined from the sum of the absolute differences between the gray levels of the pixels of said set and the gray levels of the initial set determined.

The invention also relates to a motor vehicle comprising a device as presented above.

Other features and advantages of the invention will become apparent from the following description given with reference to the appended figures given by way of non-limiting examples and in which identical references are given to similar objects.

- Figure 1 schematically illustrates a vehicle according to the invention. - Figure 2 schematically illustrates a first image and a second image, consecutive to the first image, acquired by the camera.

FIG. 3 schematically illustrates two projections of an area of pixels of a first image in a second image by homography on a projection plane located respectively at a predetermined minimum distance and at a maximum distance.

FIG. 4 schematically illustrates schematically a projection by homography, on a projection plane situated at a predetermined distance, of a set of points of a first image in a second image, the correspondence between the two images being realized via a rotation matrix and a translation matrix.

FIG. 5 schematically illustrates an epipolar segment obtained in a second image by projecting a set of pixels of a first image onto a plane consecutively at a predetermined plurality of distances between a predetermined minimum distance and a maximum distance predetermined.

FIG. 6 schematically illustrates a graph of correlation scores obtained for a pixel and its neighborhood projected along the segment of FIG. 5.

FIG. 7 diagrammatically illustrates an embodiment of the method according to the invention.

The image processing device according to the invention will now be described with reference to FIGS. 1 to 6.

With reference to FIG. 1, the image processing device 10 is mounted in a motor vehicle 1. By the term "motor vehicle" is meant, in a standard manner, a road vehicle driven by an internal combustion engine, electric or gas turbine engine or a hybrid engine such as, for example, a car, a van, a truck, etc.

The device 10 comprises a camera 100 and an image processing module 200. Preferably, the camera 100 is mounted at the upper central part of the front windshield (not shown) of the vehicle 1. Note however that the camera 100 could be mounted at any other suitable place of the vehicle 1 (side, rear etc.). The camera 100 is configured to acquire a plurality of images of the environment 2 of the vehicle 1, for example the road 2A, so that the image processing module 200 exploits them.

The image processing module 200 processes the images acquired by the camera 100 to provide driving assistance to the driver of the vehicle 1. AT As an example, such image processing may consist of detecting objects in the images, such as, for example, panels, road borders, obstacles (pedestrians or other ...), in order to inform the driver respectively a speed limit, a risk of leaving the road or a risk of collision. Such image processing can also be used to measure the height of a bridge or the position of an obstacle in order to warn the driver or perform an emergency braking.

For this purpose, with reference to FIG. 2, the image processing module 200 is configured to perform a plurality of tasks on a first image 11 and a second image 12 acquired consecutively by the camera 100.

The image processing module 200 is firstly configured to determine a rotation matrix R and a translation matrix T between the first image 11 and the second image 12, for example by using the same static object represented on both images, for example the road or a bridge, or a dynamic object moving relative to the camera 100, for example another vehicle. In the latter case, it is necessary to know where the object is located in the image (for example by using a classification method known per se) and then to determine the rotation matrix and the translation matrix between the dynamic object and the camera 100, the translation matrix T being in this case weighted by a coefficient a which can be determined in known manner by image processing knowing the size of the object.

The determination of the rotation matrix R and of the translation matrix T between the first image 11 and the second image 12 may be performed by an image processing algorithm or by the use of one or more inertial sensors and / or or geographical position (for example GPS) in a manner known per se.

By way of example, in known manner, such an algorithm can consist in determining the temporal flow between the first image 11 and the second image 12 to deduce therefrom a first so-called "fundamental" matrix, to transform this fundamental matrix into a so-called matrix "Essential" from the intrinsic calibration parameters of the camera 100, to deduce a rotation matrix R and an intermediate translation matrix aT obtained at a factor "a" and correct this intermediate translation matrix aT to obtain a final translation matrix T from an odometer of the vehicle 1. Such a method being known per se, it will not be further detailed here.

Then, again with reference to FIG. 2, the image processing module 200 is configured to select a pixel PA (1) in the first image 11 for which a corresponding pixel is sought in the second image 12. By the terms " corresponding pixel "means a pixel representing the same spatial area of the environment of the vehicle 1. Still with reference to FIG. 2, the image processing module 200 is then configured to determine in the first image 11 an initial set of pixels comprising the selected pixel PA (1) and a plurality of pixels PA (2), PA (5), located in the vicinity of the selected pixel PA (1). By the term "neighborhood" is meant in a known manner PA (2) pixels, PA (5) being around the selected pixel PA (1) in the first image 11. The number of pixels of the neighborhood is four in this example for the sake of clarity, but it goes without saying that the neighborhood of the pixel PA (1) may comprise more or fewer than four pixels (more generally "n" pixels, "n" being a natural number greater than or equal to 2). Such an initial set of pixels may be a window or a portion of the first image 11, for example a window of 10 x 10 pixels. It will be noted here that the number n of pixels and the shape of the window or the image portion may vary on purpose. It will also be noted that five pixels are represented in the set as an example but that the number of pixels of the set could of course be less than or greater than five.

The image processing module 200 is configured to determine, in the second image 12, a plurality of sets of pixels PB (1), PB (5) by projection, from the rotation matrix R and the matrix defined translation T, of the initial set determined PA (1), PA (5). This first set PB (1), PB (5) comprises a first pixel PB (1) corresponding to the selected pixel PA (1) in the first image 11.

With reference to FIGS. 3 and 4, this projection is performed on a predetermined plane W located at different predetermined distances D from the optical center C1 of the camera 100 for the first image 11.

More specifically, with reference to FIG. 3, the image processing module 200 is firstly configured to determine, in the second image 12, a first set of pixels PBmin (1), PBmin (5) by projection, from the determined rotation matrix R and the translation matrix T, from the determined initial set PA (1), PA (5) on a predetermined plane W at a predetermined minimum distance Dmin. This first set PBmin (1), PBmin (5) comprises a first pixel PBmin (1) corresponding to the selected pixel PA (1) in the first image 11.

Similarly, again with reference to FIG. 3, the image processing module 200 is configured to determine, in the second image 12, a second set of pixels PBmax (1), PBmax (5) by projection, starting from the rotation matrix R and the translation matrix T determined from the initial set determined PA (1), PA (5) on the predetermined plane W located at a predetermined maximum distance Dmax. This second set PBmax (1), PBmax (5) comprises a second pixel PBmax (1) corresponding to the pixel selected in the first image 11. Referring now to FIG. 4, the determination of the projection of a pixel of the first image 11 in the second image 12 via a projection plane W corresponds to a homography of the pixel in a reference frame C1 (which corresponds to the optical center of the camera 100 during the acquisition of the first image 11) at an origin reference C2 (which corresponds to the optical center of the camera 100 during the acquisition of the second image 12) via the projection plane W located at one of the determined distances. Such a projection by homography H is carried out from rotation matrices R and translation T, a vector n ₂ normal to the projection plane W and the distance D from the plane W with respect to the optical center C1 in the following manner : H = R - (- j).

Since such a homography is known per se, it will not be further detailed here.

In theory, the minimum distance Dmin is zero and the maximum distance Dmax corresponds to infinity. In practice, it is possible to choose a minimum distance Dmin of a few meters, for example 5 meters, and a maximum distance Dmax of a few tens of meters, for example 50 or 100 meters, especially when the selected pixel PA (1) in the first image 11 represents an identified object whose distance from the camera 100 is known, for example the road scrolling in the lower part of the images acquired by the camera 100.

The projection plane W can be determined according to the nature of the selected pixel PA (1). For example, when the selected pixel PA (1) represents a zone of the road 2A on which the vehicle 1 is traveling, a plane W corresponding to the plane of the road 2A can be used as illustrated schematically in FIG. 2. This plane W is a virtual projection plane which will make it possible to estimate by homography, that is geometrically, the corresponding position of a pixel of the first image 11 in the second image 12.

Once the first pixel PBmin (1) and the second pixel PBmax (1) have been selected, the image processing module 200 is configured to determine a segment U connecting the first pixel PBmin (1) and the second pixel PBmax (1). .

Referring to Fig. 5, the image processing module 200 is then configured to select a plurality of PB ₂ (1), PB ₅ (1) points along the determined U segment. Advantageously, the points PB ₂ (1), PB ₅ (1) can be distributed along the segment U starting from the first pixel PBmin (1) by being spaced apart by a "step" step of predetermined width, for example every 0.1 pixels. The number of selected points PB ₂ (1), PB ₅ (1) on the segment U is four in this example but it goes without saying that it could be greater or less than four.

Then, the image processing module 200 is configured to calculate, for each PB ₂ (1), PB ₅ (1) point selected along the segment U, the distance, called intermediate distance D ₂ , ..., D ₅ of the predetermined projection plane W. Such a calculation can be performed in a known manner (provided that the intrinsic calibration of the camera 100 has been determined beforehand).

Then, with reference to FIG. 4, again for each PB ₂ (1), PB ₅ (1) point selected on the segment U, the image processing module 200 is configured to determine, in the second image 12, a intermediate pixel set {(PB ₂ (1), ...; PB ₂ (5)); ...; (PB ₅ (1), ...; PB ₅ (5))} by projecting the initial set determined in the first image 11 on the predetermined plane W when the latter is situated at the intermediate distance D ₂ , .. ., D ₅ calculated for said selected PB ₂ (1), PB ₅ (1) point.

For each set of pixels {{(PBmin (1), ...; PBmin (5)), (PB ₂ (1), ..., PB ₂ (5)); ...; (PB ₅ (1), ..., PB ₅ (5)); (PBmax (1), ...; PBmax (5))} associated with a point PBmin (1), PB2 (1), PB5 (1), PBmax (1) of the determined segment U, the processing module of images 200 is configured to calculate a correlation score Si, ..., S ₆ .

There are several known methods for calculating a correlation score.

A first method is based on measurements using the statistics of the gray level difference distribution between the pixels of the first image 11 and the second image 12. For example, the sum of the absolute differences or SAD for "Sum of Absolute "Differences" in English is a simple algorithm used to find a correlation between two neighborhoods. It is determined by calculating the absolute difference of the gray levels between each pixel in the original pixel set of the first image 11 and the corresponding pixel of the associated set in the second image 12. These differences are summed to create a simple measure of similarity of the block, the norm L1 of the image of the differences.

Other known methods are based on the least square error (MSE), the Mean Mean Absolute Deviation (MAD), the sum of squared errors ( SSE or "Sum of Squared Errors"), on cross-correlation measurements that exploit the scalar product or on measurements using the derivatives of the images or on ordinal measurements that are based on the order of the gray levels pixels, etc.

Once the correlation score is determined for each set of pixels {{(PBmin (1), ...; PBmin (5)), (PB ₂ (1), ...; PB ₂ (5)); ...; (PB ₅ (1), ..., PB ₅ (5)); (PBmax (1), ...; PBmax (5))} of the second image 12, each corresponding to the projection of the original set of the first image 11 via the projection plane W located at a different distance Dmin, D ₂ , D ₅ , Dmax, the image processing module 200 is configured to determine in the second image 12 the set of pixels whose correlation score Si, ..., S ₆ is the highest, the pixel of this set that corresponds to the pixel PA (1) selected in the first image 12 being the pixel that has the best chance of actually matching it. These two pixels constituting in the second image 12 the optical flow of the pixel PA (1) selected in the first image 11.

The invention will now be described in its implementation with reference to FIGS. 2 to 7.

Firstly, in a step E1, with reference to FIG. 2, the camera 100 acquires a first consecutive image 11 and a second image 12. Note that the method can be implemented by taking as a first image 11 the image whose acquisition is temporally the first (previous or previous image in the stream of images acquired by the camera) and as the second image 12, l successive image in the flow of images. In contrast, the method can be implemented by taking as a first image 11 the image whose acquisition is temporally posterior (following image in the stream of images acquired by the camera) to the second image 12 and as the second image 12 , the previous image in the flow of images. In the following example, the previous image is chosen as the first image 11 and the posterior image (i.e. successive in the flow of images) as the second image 12.

With reference to FIG. 3, the image processing module 200 then determines, in a step E2, the rotation matrix R and the translation matrix T between the first image 11 and the second image 12.

The image processing module 200 selects, in a step E3, a pixel PA (1) in the first image 11 for which a corresponding pixel in the second image 12 must be determined.

Advantageously, the image processing module 200 can firstly select a pixel, which has no optical flow in the second image 12, in a zone of interest of the first image 11, for example a side of the image representing the edge of the road when you want to warn the driver of a risk of leaving the road, the top of the image representing the sky when you want to detect a bridge to assess the height etc.

Then, with reference to FIG. 2, for the selected pixel PA (1), the image processing module 200 determines in the first image 11, in a step E4, an initial set of pixels PA (1), PA ( 5) comprising the selected pixel PA (1) and a plurality of PA (2), PA (5) pixels located near the selected pixel PA (1). In the example of FIG. 2, for the sake of clarity, only four PA (2), PA (3), PA (4) and PA (5) pixels in the vicinity of the PA (1) pixel have been selected.

With reference to FIG. 3, the image processing module 200 determines or selects (for example from a list stored in a memory zone (not shown) of the device 10) in a step E5 a minimum distance Dmin and a maximum distance Dmax between which is the spatial point represented by the selected pixel PA (1). These values Dmin and Dmax can be determined as described above or by using the history of previous images or a geometrical knowledge of the scene, such as a flat road.

In a step E6, the image processing module 200 determines a projection plane W, of the selected pixel PA (1), in the second image 12, for example a vertical plane located facing the front of the vehicle 1. Note that steps E5 and E6 can be reversed.

With reference to FIGS. 2 and 3, in a step E7, the image processing module 200 determines, in the second image 12, a first set of pixels PBmin (1), PBmin (5) by homographic projection of the set initial determined pixels PA (1), PA (5) on the W plane, considering that the latter is placed at the determined minimum distance Dmin. This first set PBmin (1), PBmin (5) comprises a corresponding first pixel PBmin (1), by homographic projection on the W plane at the minimum distance Dmin, to the selected pixel PA (1) in the first image 11.

Similarly, in a step E8, the image processing module 200 determines, in the second image 12, a second set of pixels PBmax (1), PBmax (5) by homographic projection of the initial set determined PA (1). , PA (5) on the predetermined plane W located at the predetermined maximum distance Dmax. This second set PBmax (1), PBmax (5) comprises a corresponding second pixel PBmax (1), by homographic projection of the plane W at the maximum distance Dmax, to the selected pixel PA (1) in the first image 11. It goes without saying that steps E7 and E8 can be reversed.

With reference to FIGS. 3 to 5, the image processing module 200 then determines, in a step E9, the segment U connecting the first pixel PBmin (1) and the second pixel PBmax (1) in the second image 12 and then selects in a step E10, a plurality of points PB ₂ (1), PB ₅ (1) on the segment U thus determined, for example every 0.1 pixels.

For each point PB ₂ (1), PB ₅ (1) determined, the image processing module 200 performs two successive substeps.

In a sub-step E1 1 A, the image processing module 200 calculates the distance (called the intermediate distance) D ₂ ,..., D ₅ from the projection plane W. It will be noted that this distance is already known for PBmin (1) and PBmax (1) (respectively Dmin and Dmax). To do this, we express each point PB ₂ (1), PB ₅ (1) as a function of PA (1), of the rotation matrix R and of the translation matrix T, or in simplified writing:

PB = R. PA + T

PA = the actual coordinates of points PA and PB.

The coordinates of the point PA in the first image 11 are given by: UA =

^{and V} A = f- ^where f is the focal length of the camera 100.

Likewise, the coordinates of the point PB in the second image 12 are given by:

u _B = f. - ev _B = f. -

We thus have 4 equations with 3 unknowns (X, Y, Z) which can be solved in a known way to obtain in particular the value of Z which corresponds to the sought distance (D ₂ , ..., D ₅ ).

In a substep E1 1 B, the image processing module 200 then determines in the second image 12 an intermediate set of pixels {(PB ₂ (1), ...; PB ₂ (5)); ...; (PB ₅ (1), ...; PB ₅ (5))} by projecting the initial set determined PA (1), PA (5) on the plane W located at the intermediate distance D ₂ , ... , D ₅ calculated.

Then, as illustrated in FIG. 6, in a step E12, the image processing module 200 calculates a correlation score S ₂ , S ₅ for the intermediate pixel set as well as a correlation score Si for the first set of pixels PBmin (1), PBmin (5) and a correlation score Se for the second set of pixels PBmax (1), PBmax (5).

In a step E13, the image processing module 200 finally determines, in the second image 12, the set of pixels with the highest correlation score (the score S ₃ in this example). The pixel of this set corresponding to the projection of the pixel PA (1) constitutes the pixel which has the best correspondence with the pixel PA (1) selected in the first image 11, in terms of representation of the same spatial area of the environment of the vehicle 1.

Note that a sub-pixellic refinement can be achieved to obtain a better accuracy by taking into account neighboring scores.

For dynamic objects, rotation and translation between images is expressed relative to a world reference that is related to the dynamic object.

The method according to the invention thus makes it possible to easily and efficiently detect objects in the images captured by the camera 100 in order to assist the driver in his driving of the vehicle 1. It is furthermore specified that the present invention is not limited to the examples described above and is capable of numerous variants accessible to those skilled in the art.

Claims

1. Method for processing images (11, 12) acquired by a camera (100) of a motor vehicle (1), said method comprising the steps of:

• acquisition (E1) of a first image (11) and a second image (12) by the camera (100),

· determination (E2) of a rotation matrix (R) and a translation matrix (T) between the first image and the second image,

• selection (E3) of at least one pixel (PA(1)) in the first image (11), said method being characterized in that it comprises the following steps:

• for each selected pixel (PA(1)) in the first image (11):

- determination (E4), in the first image (11), of an initial set of pixels (PA(1), PA(5)) comprising the selected pixel (PA(1)) and a plurality of pixels (PA( 2), PA(5)) located in the vicinity of the selected pixel (PA(1)),

- determination (E7), in the second image (12), of a first set of pixels (PBmin(1), PBmin(5)) by projection, onto a predetermined plane (W) located at a minimum distance (Dmin) predetermined, from the determined initial set (PA(1), PA(5)), from the rotation matrix (R) and the translation matrix (T) determined, said first set (PBmin(1), PBmin(5)) comprising a first pixel (PBmin(1)) corresponding to the selected pixel (PA(1)) in the first image (11),

- determination (E8), in the second image (12), of a second set of pixels (PBmax(1), PBmax(5)) by projection onto the predetermined plane (W) located at a predetermined maximum distance (Dmax) , from the determined initial set (PA(1), PA(5)), from the rotation matrix (R) and the translation matrix (T) determined, said second set (PBmax(1), PBmax (5)) comprising a second pixel (PBmax(1)) corresponding to the selected pixel (PA(1)) in the first image (11),

- determination (E9) of a segment (U) connecting the first pixel (PBmin(1)) and the second pixel (PBmax(1)),

- selection (E10) of a plurality of points ((PB ₂ (1),..., PB ₅ (1)) on the determined segment (U), - for each point ((PB ₂ (1),..., PB ₅ (1 )) selected:

• calculation (E1 1 A) of the distance (D ₂ , ... D ₅ ) of the predetermined projection plane (W),

• determination (E1 1 B), in the second image (12), of a set of intermediate pixels {(PB ₂ (1), ...; PB ₂ (5)); ... ; (PB ₅ (1), ... ;

PB ₅ (5))} by projection onto the predetermined plane (W) located at the calculated distance (D ₂ , ... D ₅ ), from the determined initial set (PA(1), PA(5)), from the rotation matrix (R) and the translation matrix (T) determined,

- for each set of pixels {{(PBmin(1), ...; PBmin(5)), (PB ₂ (1), ... ;

PB ₂ (5)); ... ; (PB ₅ (1), ...; PB ₅ (5)); (PBmax(1), ...; PBmax(5))} associated with a point (PBmin(1), PB ₂ (1), PB ₅ (1), PBmax(1)) of the determined segment (U), calculation (E12) of a correlation score (Si,...,S ₆ ),

- determination (E13) in the second image (12), from the correlation score (Si,...,S ₆ ) calculated, of the pixel corresponding to the same object on the first image (11) and on the second image (12).

2. Method according to claim 1, characterized in that it further comprises a step (E5) of determining the minimum distance (Dmin) and the maximum distance (Dmax).

3. Method according to one of claims 1 and 2, characterized in that it further comprises a step (E6) of determining the projection plane (W).

4. Method according to any one of claims 1 to 3, characterized in that the correlation score (Si,...,S ₆ ) of a set of pixels {{(PBmin(1), ...; PBmin(5)), (PB ₂ (1), ...; PB ₂ (5)); ... ; (PB ₅ (1), ...; PB ₅ (5)); (PBmax(1), ...; PBmax(5))} associated with a point (PBmin(1), PB ₂ (1), PB ₅ (1), PBmax(1)) of the determined segment (U) is calculated from the sum of the absolute differences between the gray levels of the pixels of said set and the gray levels of the determined initial set (PA(1), PA(5)).

5. Image processing device (10) for a motor vehicle (1), said device (10) comprising a camera (100) capable of acquiring a first image (11) and a second image (12) and a processing module of images (200) configured for:

• determine a rotation matrix (R) and a translation matrix (T) between the first image (11) and the second image (12) acquired by the camera (100),

• select at least one pixel (PA(1)) in the first image (11), said device being characterized in that the processing module is further configured to:

• for each pixel (PA(1)) selected in the first image (11):

- determine, in the first image (11), an initial set of pixels (PA(1), PA(5)) comprising the selected pixel (PA(1)) and a plurality of pixels (PA(2), PA( 5)) located in the vicinity of the selected pixel (PA(1)),

- determine, in the second image (12), a first set of pixels (PBmin(1), PBmin(5)) by projection, from the rotation matrix (R) and the translation matrix (T) determined , of the determined initial set (PA(1), PA(5)) on a predetermined plane (W) located at a predetermined minimum distance (Dmin), said first set (PB(1), PB(5)) comprising a first pixel (PB(1)) corresponding to the selected pixel (PA(1)) in the first image (11), - determine, in the second image (12), a second set of pixels (PBmax(1), PBmax (5)) by projection, from the rotation matrix (R) and the translation matrix (T) determined, of the determined initial set (PA(1), PA(5)) onto the plane (W ) predetermined located at a predetermined maximum distance (Dmax), said second set (PBmax(1), PBmax(5)) comprising a second pixel (PBmax(1)) corresponding to the selected pixel (PB(1)) in the first image (11),

- determine a segment (U) connecting the first pixel (PBmin(1)) and the second pixel (PBmax(1)),

- select a plurality of points ((PB ₂ (1),..., PB ₅ (1)) on the determined segment (U),

- for each point ((PB ₂ (1),..., PB ₅ (1 )) selected:

• calculate the distance (D ₂ , ... D ₅ ), called intermediate distance, of the predetermined projection plane (W), · determine, in the second image (12), an intermediate set of pixels {(PB ₂ (1 ), ...; PB ₂ (5)); ... ; (PB ₅ (1), ...; PB ₅ (5))} by projection, from the determined rotation matrix (R) and translation matrix (T), of the determined initial set (PA (1), PA(5)) on the predetermined plane (W) located at the calculated intermediate distance (D ₂ , ... D ₅ ),

- for each set of pixels {{(PBmin(1), ...; PBmin(5)), (PB ₂ (1), ... ;

PB ₂ (5)); ... ; (PB ₅ (1), ...; PB ₅ (5)); (PBmax(1), ...; PBmax(5))} associated with a point (PBmin(1), PB ₂ (1), PB ₅ (1), PBmax(1)) of the determined segment (U), calculate a correlation score (Si,...,S ₆ ),

- determine in the second image (12), from the correlation score (Si,...,S ₆ ) calculated, the pixel corresponding to the same object on the first image (11) and on the second image (12) .

6. Device (10) according to claim 5, characterized in that the image processing module (200) is configured to determine the minimum distance (Dmin) and the maximum distance (Dmax).

7. Device (10) according to one of claims 5 and 6, characterized in that the image processing module (200) is configured to determine the projection plane (W).

8. Device (10) according to any one of claims 5 to 7, characterized in that the image processing module (200) is configured to calculate the correlation score (Si,...,S ₆ ) d 'a set of pixels {{(PBmin(1), ...; PBmin(5)), (PB ₂ (1), ...; PB ₂ (5)); ... ; (PB ₅ (1), ...; PB ₅ (5)); (PBmax(1), ...; PBmax(5))} associated with a point (PBmin(1), PB ₂ (1), PB ₅ (1), PBmax(1)) of the segment (U) determined at from the sum of the absolute differences between the gray levels of the pixels of said set and the gray levels of the determined initial set (PA(1), PA(5)).

9. Motor vehicle (1), characterized in that it comprises a device (10) according to any one of claims 5 to 8.